On Fri, Apr 24, 2020 at 06:00:50PM +0530, Syed Nayyar Waris wrote:
> This patch reimplements the thunderx_gpio_set_multiple function in
> drivers/gpio/gpio-thunderx.c to use the new for_each_set_clump macro.
> Instead of looping for each bank in thunderx_gpio_set_multiple
> function, now we can skip bank which is not set and save cycles.
>
> Cc: Robert Richter <[email protected]>
> Cc: Linus Walleij <[email protected]>
> Cc: Bartosz Golaszewski <[email protected]>
> Signed-off-by: Syed Nayyar Waris <[email protected]>
> Signed-off-by: William Breathitt Gray <[email protected]>
> ---
> drivers/gpio/gpio-thunderx.c | 12 ++++++++----
> 1 file changed, 8 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpio/gpio-thunderx.c b/drivers/gpio/gpio-thunderx.c
> index 9f66dea..74aea25 100644
> --- a/drivers/gpio/gpio-thunderx.c
> +++ b/drivers/gpio/gpio-thunderx.c
> @@ -275,12 +275,16 @@ static void thunderx_gpio_set_multiple(struct gpio_chip *chip,
> unsigned long *bits)
> {
> int bank;
> - u64 set_bits, clear_bits;
> + u64 set_bits, clear_bits, gpio_mask;
> + const unsigned long bank_size = 64;
> + unsigned long offset;
> +
> struct thunderx_gpio *txgpio = gpiochip_get_data(chip);
>
> - for (bank = 0; bank <= chip->ngpio / 64; bank++) {
> - set_bits = bits[bank] & mask[bank];
> - clear_bits = ~bits[bank] & mask[bank];
> + for_each_set_clump(offset, gpio_mask, mask, chip->ngpio, bank_size) {
> + bank = offset / bank_size;
> + set_bits = bits[bank] & gpio_mask;
> + clear_bits = ~bits[bank] & gpio_mask;
> writeq(set_bits, txgpio->register_base + (bank * GPIO_2ND_BANK) + GPIO_TX_SET);
> writeq(clear_bits, txgpio->register_base + (bank * GPIO_2ND_BANK) + GPIO_TX_CLR);
> }
> --
> 2.7.4
We noticed in the original code that this set_multiple callback does not
appear to work correctly on systems where BITS_PER_LONG == 32. On those
systems, the bits and mask values are 32-bit, but the for loop jumps 64
bits at a time -- that means the loop is skipping the upper 32 gpio
lines of every iteration.
Is the gpio-thunderx driver only intended for 64-bit systems? Or this
behavior a bug?
William Breathitt Gray
On Fri, Apr 24, 2020 at 09:55:47AM -0400, William Breathitt Gray wrote:
> On Fri, Apr 24, 2020 at 06:00:50PM +0530, Syed Nayyar Waris wrote:
> > This patch reimplements the thunderx_gpio_set_multiple function in
> > drivers/gpio/gpio-thunderx.c to use the new for_each_set_clump macro.
> > Instead of looping for each bank in thunderx_gpio_set_multiple
> > function, now we can skip bank which is not set and save cycles.
> >
> > Cc: Robert Richter <[email protected]>
> > Cc: Linus Walleij <[email protected]>
> > Cc: Bartosz Golaszewski <[email protected]>
> > Signed-off-by: Syed Nayyar Waris <[email protected]>
> > Signed-off-by: William Breathitt Gray <[email protected]>
> > ---
> > drivers/gpio/gpio-thunderx.c | 12 ++++++++----
> > 1 file changed, 8 insertions(+), 4 deletions(-)
> >
> > diff --git a/drivers/gpio/gpio-thunderx.c b/drivers/gpio/gpio-thunderx.c
> > index 9f66dea..74aea25 100644
> > --- a/drivers/gpio/gpio-thunderx.c
> > +++ b/drivers/gpio/gpio-thunderx.c
> > @@ -275,12 +275,16 @@ static void thunderx_gpio_set_multiple(struct gpio_chip *chip,
> > unsigned long *bits)
> > {
> > int bank;
> > - u64 set_bits, clear_bits;
> > + u64 set_bits, clear_bits, gpio_mask;
> > + const unsigned long bank_size = 64;
> > + unsigned long offset;
> > +
> > struct thunderx_gpio *txgpio = gpiochip_get_data(chip);
> >
> > - for (bank = 0; bank <= chip->ngpio / 64; bank++) {
> > - set_bits = bits[bank] & mask[bank];
> > - clear_bits = ~bits[bank] & mask[bank];
> > + for_each_set_clump(offset, gpio_mask, mask, chip->ngpio, bank_size) {
> > + bank = offset / bank_size;
> > + set_bits = bits[bank] & gpio_mask;
> > + clear_bits = ~bits[bank] & gpio_mask;
> > writeq(set_bits, txgpio->register_base + (bank * GPIO_2ND_BANK) + GPIO_TX_SET);
> > writeq(clear_bits, txgpio->register_base + (bank * GPIO_2ND_BANK) + GPIO_TX_CLR);
> > }
> > --
> > 2.7.4
>
> We noticed in the original code that this set_multiple callback does not
> appear to work correctly on systems where BITS_PER_LONG == 32. On those
> systems, the bits and mask values are 32-bit, but the for loop jumps 64
> bits at a time -- that means the loop is skipping the upper 32 gpio
> lines of every iteration.
>
> Is the gpio-thunderx driver only intended for 64-bit systems? Or this
> behavior a bug?
>
> William Breathitt Gray
Nevermind, I spoke too soon; I see there's a depends on 64BIT line in
the Kconfig for GPIO_THUNDERX.
William Breathitt Gray